基于深度学习的残缺冲击波信号构建方法

Reconstruction Method of Incomplete Shock Wave Signals Based on Deep Learning

爆炸冲击波参数是评估弹药威力的主要依据之一,而在实际测试过程中,测试系统可能受破片或其他因素影响而导致损坏,从而无法捕获完整信号,影响了后续的毁伤评估。本文针对该问题提出了一种基于双向长短时记忆网络(BiLSTM)融合多头自注意力模块的方法对残缺的冲击波信号进行完整性构建,采用BiLSTM分析了冲击波信号的局部时序依赖关系,以多头自注意力模块捕捉信号中的频率信息,最终实现了时序信号与频率信息的融合,从而得到完整的冲击波信号。在一次信息采集过程中,所测得的信号数据通常只有数十组,从而导致了小样本问题,本文建立了以LSTM单元为生成器的GAN网络,对完整的冲击波信号进行扩充,增强了数据集容量。基于扩充数据集的构建实验结果表明,本文所提方法构建的完整信号与原始信号之间的MSE和MAE分别为0.0068和0.1462,优于LSTM、BiLSTM、CNN+BiLSTM等方法,本文所提方法可以满足构建残缺冲击波信号的实际需求。

The parameters of explosion shock waves are one of the main criteria for evaluating the power of ammunition.However,during the actual testing process,the testing system may be damaged by fragments or other factors,making it unable to capture the complete signal and thus affecting subsequent dam‐age assessment.Therefore,this article proposed a method based on bidirectional long short-term memory network(BiLSTM)and multi head self-attention module fusion to construct the integrity of incomplete shockwave signals.BiLSTM was used to analyze the local temporal dependencies of shockwave signals,and the multi head self-attention module captured frequency information in the signal.Finally,the fusion of temporal signals and frequency information was achieved,resulting in a complete shockwave signal.In the process of information collection,the measured signal data is usually only dozens of sets,which leads to the problem of small sample size.This article established a GAN network with LSTM units as the gen‐erator to expand the complete shock wave signal and enhance the dataset capacity.The experimental results based on the expanded dataset show that the MSE and MAE between the complete signal constructed by the method proposed in this paper and the original signal are 0.0068 and 0.1462,respectively,which are superior to the LSTM,BiLSTM,and CNN+BiLSTM methods.The method pro‐posed in this paper meets the practical requirements for constructing incomplete shock wave signals.